Simplify User DataSource

[mdr223]

Writing a custom user DataSource is too difficult in PZ. It is made especially more complicated in the context of validation DataSources which are needed for sentinel (optimized) execution.

Users should only need to override a single get_item (or __iterator__) method which yields outputs from the DataSource. In my opinion, users should not need to know how to construct (or use) PZ's DataRecord class.

Instead, my proposal is the following:

• users implement the get_item method, which takes an integer idx and returns a dict
• the dict keys are the field names for the record to be returned (which must match those in the DataSource schema); the values are the values for those fields
• PZ takes the output dict and converts it into a DataRecord on behalf of the user

One further step may include eliminating the need for users to specify a Schema at all; I will explore this option as part of the work on this issue.